Experiments with Detector-based Conditional Random Fields in Phonetic Recogntion

1. Experiments with Detector-based Conditional Random Fields in Phonetic Recogntion Jeremy Morris & Eric Fosler-Lussier 04/19/2007

2. Outline • Background • Previous Work • Feature Combination Experiments • Viterbi Realignment Experiments • Conclusions and Future Work

3. Background • Goal: Integrate outputs of speech attribute detectors together for recognition • e.g. Phone classifiers, phonological feature classifiers • Attribute detector outputs highly correlated • Stop detector vs. phone classifier for /t/ or /d/ • Accounting for correlations in HMM • Ignore them (decreased performance) • Full covariance matrices (increased parameters) • Explicit decorrelation (e.g. Karhunen-Loeve transform)

4. Background • Conditional Random Fields (CRFs) • Discriminative probabalistic model • Used successfully in various domains such as part of speech tagging and named entity recogntion • Directly defines a posterior probability of a sequence Y given an input sequence X • e.g. P(Y|X) • Does not make independence assumptions about correlations among input features

5. Background • CRFs for ASR • Phone Classification (Gunawardana et al., 2005) • Uses sufficient statistics to define feature functions • Different approach than NLP tasks using CRFs • Define binary feature functions to characterize observations • Our approach follows the latter method • Use neural networks to provide “soft binary” feature functions (e.g. posterior phone outputs)

6. Conditional Random Fields /k/ /k/ /iy/ /iy/ /iy/ • Based on the framework of Markov Random Fields

7. /k/ /k/ /iy/ /iy/ /iy/ X X X X X Conditional Random Fields • Based on the framework of Markov Random Fields • A CRF iff the graph of the label sequence is an MRF when conditioned on a set of input observations (Lafferty et al., 2001)

8. /k/ /k/ /iy/ /iy/ /iy/ X X X X X Conditional Random Fields • Based on the framework of Markov Random Fields • A CRF iff the graph of the label sequence is an MRF when conditioned on the input observations State functions help determine the identity of the state

9. /k/ /k/ /iy/ /iy/ /iy/ X X X X X Transition functions add associations between transitions from one label to another Conditional Random Fields • Based on the framework of Markov Random Fields • A CRF iff the graph of the label sequence is an MRF when conditioned on the input observations State functions help determine the identity of the state

10. Conditional Random Fields • CRF defined by a weighted sum of state and transition functions • Both types of functions can be defined to incorporate observed inputs • Weights are trained by maximizing the likelihood function via gradient descent methods

11. Previous Work • Implemented CRF models on data from phonetic attribute detectors • Performed phone recognition • Compared results to Tandem/HMM system on same data • Experimental Data • TIMIT corpus of read speech

12. Attribute Selection • Attribute Detectors • ICSI QuickNet Neural Networks • Two different types of attributes • Phonological feature detectors • Place, Manner, Voicing, Vowel Height, Backness, etc. • N-ary features in eight different classes • Posterior outputs -- P(Place=dental | X) • Phone detectors • Neural networks output based on the phone labels • Trained using PLP 12+deltas

13. Experimental Setup • CRF code • Built on the Java CRF toolkit from Sourceforge • http://crf.sourceforge.net • Performs maximum log-likelihood training • Uses Limited Memory BGFS algorithm to perform minimization of the log-likelihood gradient

14. Experimental Setup • Feature functions built using the neural net output • Each attribute/label combination gives one feature function • Phone class: s/t/,/t/ or s/t/,/s/ • Feature class: s/t/,stop or s/t/,dental

15. Experimental Setup • Baseline system for comparison • Tandem/HMM baseline (Hermansky et al., 2000) • Use outputs from neural networks as inputs to gaussian-based HMM system • Built using HTK HMM toolkit • Linear inputs • Better performance for Tandem with linear outputs from neural network • Decorrelated using a Karhunen-Loeve (KL) transform

16. Initial Results (Morris & Fosler-Lussier, 06) * Significantly (p>0.05) better than comparable Tandem monophone system * Significantly (p>0.05) better than comparable CRF monophone system

17. Feature Combinations • CRF model supposedly robust to highly correlated features • Makes no assumptions about feature independence • Tested this claim with combinations of correlated features • Phone class outputs + Phono. Feature outputs • Posterior outputs + transformed linear outputs • Also tested whether linear, decorrelated outputs improve CRF performance

18. Feature Combinations - Results * Significantly (p>0.05) better than comparable posterior or linear KL systems

19. Viterbi Realignment • Hypothesis: CRF results obtained by using only pre-defined boundaries • HMM allows “boundaries” to shift during training • Basic CRF training process does not • Modify training to allow for better boundaries • Train CRF with fixed boundaries • Force align training labels using CRF • Adapt CRF weights using new boundaries

20. Viterbi Realignment - Results * Significantly (p>0.05) better than comparable CRF monophone system * Significantly (p>0.05) better than comparable Tandem 4mix triphone system * Signficantly (p>0.05) better than comparable Tandem 16mix triphone system

21. Conclusions • Using correlated features in the CRF model did not degrade performance • Extra features improved performance for the CRF model across the board • Viterbi realignment training significantly improved CRF results • Improvement did not occur when best HMM-aligned transcript was used for training

22. Future Work • Recently implemented stochastic gradient training for CRFs • Faster training, improved results • Work currently being done to extend the model to word recognition • Also examining the use of transition functions that use the observation data